Generative Diffusion for Inverse Materials Design (2D)

AI-driven materials discovery needs a generative model that proposes new candidate crystal structures, the way GNoME-style pipelines search chemical space for stable compounds. This 2D simulator renders the same simplified score-based diffusion sampler as its 3D sibling: atoms begin as pure positional and chemical noise around a chosen planar lattice (square, centered square or hexagonal), then iteratively denoise toward an arrangement set by a cluster-expansion pair-interaction energy. A negative interaction constant J guides the sampler toward an ordered compound, a positive one toward phase-separated clusters, and a dopant-fraction control sets the target composition — with live formation-energy and crystalline-order readouts, a rolling energy/order sparkline, and a pannable, zoomable view tracking the structure as it condenses out of noise.